Region Detector Evaluation Data Set

The Evaluation is focused on affine invariant region detectors, e.g. :

• Image Sequences with changes in Illumination, Rotation, Perspective, and Scale used to evaluate the accuracy of region detectors
• Highly-accurate homography matrices H for each image pair. The matrices determine the mapping between the images. The matrices are generated without using any feature detector.

• Use matlab code provided by Mikolajczyk et al. for detector evaluation

Examples: Perspective Change

• Projection of Scene Point P to Camera Planes:

• Scene restricted to planar geometry
• With homogeneous coordinates: H*p=p'

• GRACE Sequence (1536x1024) [viewpoint change]
  

• UNDERGROUND Sequence (1536x1024) [viewpoint change + pattern repetitions]
  

• COLORS Sequence (1536x1024) [viewpoint change + pattern repetitions]
  

• POSTERS Sequence (1536x1024) [viewpoint change + texture repetitions]
  

• THERE Sequence (1536x1024) [strong viewpoint change]
  

Download Highly-Accurate Data Sets

• Our data sets overcome the problem of limited accuracy of the data provided by Mikolajczyk et al.
• See SSCI2011 Paper for details.
• We provide
• Homographies with increased accuracy for Mikolajczyk's image sequences
• for Bark : BarkHomAcc
• for Graffiti : GrafHomAcc (modified images included)
• for Bikes : BikesHomAcc
• New, highly accurate data sets (Images + Homographies)
• Grace
• Underground
• Colors
• Posters
• There
• If you use the data for evaluation, please cite our paper

Publications

Conference Contributions
	Kai Cordes, Bodo Rosenhahn, Jörn Ostermann Increasing the Accuracy of Feature Evaluation Benchmarks Using Differential Evolution IEEE Symposium Series on Computational Intelligence (SSCI) - IEEE Symposium on Differential Evolution (SDE), IEEE Computer Society, Paris, France, April 2011 (pdf, link) BibTeX (paper page)